Assessment of Different Methods for Pediatric Meningitis Dosing Clinical Decision Support

Abstract

Background: Indication-specific medication dosing support is needed to improve pediatric dosing support.

Objective: To compare the sensitivity and positive predictive value (PPV) of different meningitis dosing alert triggers and dosing error rates between antimicrobials with and without meningitis order sentences.

Methods: We retrospectively analyzed 4-months of pediatric orders for antimicrobials with meningitis-specific dosing. At the time of the order, it was determined if the antimicrobial was for meningitis management, if a cerebrospinal fluid (CSF) culture was ordered, and if a natural language processing (NLP) system could detect "meningitis" in clinical notes.

Results: Of 1383 orders, 243 were for the management of meningitis. A CSF culture or NLP combination trigger searching the electronic health record since admission yielded the greatest sensitivity for detecting meningitis management (67.5%, P < 0.01 vs others), but dosing error detection was similar if the trigger only searched 48 hours preceding the order (68.8% vs 62.5%, P = 0.125). Using a CSF culture alone and a 48-hour time frame had a higher PPV versus a combination with a 48-hour time frame (97.1% vs 80.9%, P < 0.001), and both triggers had a higher PPV than others ( P < 0.001). Antimicrobials with meningitis order sentences had fewer dosing errors (19.8% vs 43.2%, P < 0.01). Conclusion and Relevance: A meningitis dosing alert triggered by a combination of a CSF culture or NLP system and a 48-hour triggering time frame could provide reasonable sensitivity and PPV for meningitis dosing errors. Order sentences with indication-specific recommendations may provide additional dosing support, but additional studies are needed.

Keywords: clinical decision support; dosing alerts; information technology; medical informatics; natural language processing; pediatric.